An Empirical Study of Error Characteristics in Audit Populations

Abstract

Most modeling and analysis of audit populations has been based on surprisingly little formal study of actual empirical evidence. The single major empirical investigation to date has been the pioneering work of Neter and Loebbecke [1975; 1977]. This study examined the behavior of many statistical procedures commonly used in auditing, based on simulated audit populations constructed by extrapolating characteristics of errors found in audits of four actual populations. Neter and Loebbecke made a thorough examination of the four audit populations and reported on both the shapes of the book value distributions and error characteristics including error rate, magnitude, and changes in both rate and magnitude as a function of book value. In spite of their limited number, these four populations represented considerable variety in terms of type of account, degree of skewness, error rate, and balance of errors between underand overstatements. The effective size of the empirical data base was greatly expanded by the authors' careful control of these error characteristics in creating simulated populations with varying error rates. A major conclusion of the study was that a number of widely used statistical procedures, particularly those adopted from the sample survey literature, can be seriously misleading in the auditing context. This conclusion concurs with that of Kaplan [1973], who based his analysis on